An Improved Design for Neural-Network-Based Model Predictive Control of Three-Phase Inverters

In recent days the voltage source inverters (VSI) controlled with several conventional control techniques usually contribute to variable switching frequency, as this control techniques produce tuning on the basis of ideal voltage status chosen during each sampling time. Variable switching frequencies create it hard to construct voltage source inverter input filters. As there is a great demand for the modern control techniques need for the contemporary 3-phase inverter tuning techniques with an input LC filter, that can generates quality and less presence of total harmonic distortion (THD) are required. This paper introduces improved scheme for a two-stage converter relies on the combination of model predictive controller (MPC) and improved feed-forward neural network, in order to reduce THD and improve the system's continuous and dynamic performance for various kinds of load. First, in the training stage, MPC is used as an expert to produce information necessary to train the proposed neural network. That can have perfect tuning, for effective online training without using MPC for voltage monitoring purposes. The improved control strategy is simulated using the MATLAB/ Simulink tool, carrying different load conditions into consideration.